Network Coding in P2P-Systems Christian Ortolf

Overview ● Introduction ● Galois fields ● Encoding/Decoding of Files ● Gain – Coupon Collector's problem ● Problems ● Security ● Other fields where it can be used ● Research in Freiburg

Introduction - History ● 1999 first time used R.W.Yeung and Z. Zhang, "Distributed Source Coding for Satellite Communications" ● 2000 Definition of Network Coding, Max-Flow Min-Cut Theorem in Ahlswede, R., Cai, N., Li, S.Y.R., Yeung, R.W.: Network information flow. ● 2005 through Avalanche, Network Coding gets into the media, "Network Coding for Large Scale Content Distribution", C. Gkantsidis, P. Rodriguez ● Since 2000 more than 200 Papers dealing with Network Coding

Introduction – Canonical Example ● Max-Flow – Min-Cut Theorem ● Max Flow can't be reached here without coding

Galois field – Évariste Galois ● * 25. Oktober 1811 in Bourg-la-Reine ● † 31. Mai 1832 in Paris

Galois field ● Field ● Notation ● Example: M = {0,1} with + and * defined as: + | 0 1 · | | | | | 0 1 ● Construction: Irreducible Polynomes

Galois Field – Example

Encoding of files ● From "Network Coding for Large Scale Content Distribution", C. Gkantsidis, P. Rodriguez

Encoding - Example

Decoding of Files ● Check if Coefficient vector is Helpful ● Download up to N Blocks ● Invert Matrix and Decode

Gain – Solves Coupon Collector's Problem ● Easier Protocol – Less Knowledge Exposed – Better Inscentive mechanisms possible ● High Chance to finish File if source leaves

Gain ● From "Network Coding for Large Scale Content Distribution", C. Gkantsidis, P. Rodriguez

Problems ● Small Blocks make large Coefficient Vectors – 512KB blocks on 4GB file with GF( ) makes 64MB ● CPU usage – Encoding of aBlockO(n) – DecodingO(n²) – InvertingO(n³)

Problems - CPU ● From Anatomy of a P2P Content Distribution system Gkantsidis et al ● Average: 1.4Mbit down 300Kbit/s up ● On 2GHZ Pentium 4 and 512MB ram

Security ● Usual Hashfunctions don't work – Homomorphic Hashfunctions as solution ● Very slow – Cooperative hashing ● Complex ● Needs SRCs to protect against DoS – Bulk Hashing – Secure Randomn Checksums (SRCs) ● Harder to get info from a tapped wire ● Datainjection even without Hashing nearly impossible

Other Fields where it can be used ● In Multicast Networks (example from NC Primer) ● Reduce Cost in WSN / Storage redundancy ● Network Tomography ● P2P - SAN

Research in Freiburg - TooFree ● Finding solutions for the too large CPU cost

Thank you for your Attention Questions?